This study used recursive partitioning to identify thresholds of social influences that predict lifetime smoking status among adolescents. It analyzed data from over 1,000 students followed from 6th to 12th grade. The analysis identified 13 groups defined by combinations of social influence variables (number of smoker friends, social norms, smoker parents) and demographics that accurately predicted 76.5% of students' smoking status. Key findings were that having 2 or more smoker friends in high school or any smoker friends in junior high strongly increased the probability of becoming a smoker, while having no smoker friends or parents decreased the likelihood. The study implications are that interventions should target preventing smoking influences starting in junior high and focus on students with 2 or more smoker friends.

1. Identifying Threshold of Social Influences
on Lifetime Smoking Status among
Adolescents
– A Recursive Partitioning Approach
Yue Liao, MPH
Jimi Huh, PhD
Zhaoqing Huang, MD, MA
Arif Ansari, PhD
Mary Ann Pentz, PhD
Chih-Ping Chou, PhD
Presented at the 33rd Annual Meeting of the Society of Behavioral Medicine
April, 2012
Contact: yueliao@usc.edu

2. Social Influences and Cigarette
Smoking
• Adolescents’ cigarette smoking behavior is
affected by the ones of their friends and
parent
• peer influences are believed to be the most
significant psychosocial risk factors for
cigarette smoking
• parental behavior may have different degree
of influence on child at different ages
Avenevoli & Merikangas, 2003; Kobus, 2003; Hoffman et al., 2006; Darling & Cumsille, 2003
2

3. • Combined effects of peer and parental
influences
• Non-smoking parents had a buffering effects
on peer influences
Li, Pentz, & Chou, 2002
3

4. • However, the threshold of such effects
have not yet been well examined
• i.e., how many smoker friends one “needs” to
have to be considered as a substantial risk
factor
• Combinations of peer and parental
influences with different thresholds?
• May be useful to identify high-risk groups
4

5. Current Study
• To identify combinations and thresholds of
social influences variables that predict
lifetime smoking status among
adolescents
5

6. Participants
• 1,073 students from the Midwestern
Prevention Project
• a longitudinal study that followed participants
yearly from 6th/7th to 12th grade
• Students were from Indianapolis, IN
• 48.7% male, 76.0% Caucasian
• 33.6% from low socioeconomic status family
• 74.8% from public schools
• 48.6% in the intervention group
6

7. Social Influence Variables
• Peer influences
• perceived friend use (1-7)
• “How many of your close friends use cigarettes?”
• perceived social norms (1-10)
• “Out of every 100 students in your age, how many do you
think smoke cigarettes at least once a month?”
• Parental influences
• perceived parent use (0-2)
• “How many of the two important adults in your life use
cigarettes”
7

8. • Responses from 6th/7th to 8th grade were
averaged to represent social influence
during middle/junior high school (JHS)
period
• Responses from 9th to 12th grade were
averaged to represent high school (HS)
period
8

9. Lifetime Smoking Status
• Students were considered as “lifetime non-
smokers” if
• selected “none” or “one puff to one cigarette”
to the question
• “How many cigarettes have you smoked in your
whole life?”
• at each wave of the surveys
• At 12th grade, 29.7% of the students were
identified as “lifetime non-smokers”
9

10. Statistical Methods
• Recursive partitioning was used to classify
membership (lifetime smokers vs. non-
smokers) based on social influences &
demographic variables
• a binary classification method that creates a
decision tree
• can examine the effects of combination of
multiple predictors
• if a person has x, y, and z, what is the probability
of having condition q
10

11. • Combination of the predictors and the
associated cut-point was selected based
on conditional probability that can
minimize the entropy (randomness) in the
model
• Analysis was performed using JMP 9.0.0
11

12. Results
• 13 groups with different combinations of
social influences and demographic
variables that distinguish between lifetime
smokers vs. non-smokers were identified
• Accuracy rate of predicting smokers vs.
non-smokers was 76.5%
12

13. 13

14. 14

15. Combinations of factors that predict lifetime Probabilit
smokers y
1. Have >=8 smoker friends during HS (N=124) 96.57%
2. Have 3-7 smoker friends during HS + White + No 95.96%
intervention (N=55)
3. Have 2 smoker friends during HS + >=1 smoker friends 92.27%
during JHS + Non-White (N=16)
Combinations of factors that predict lifetime
non-smokers
1. Have no smoker friends during JHS + no smoker friends 77.06%
during HS + <2 smoker parent (N=54)
2. Have no smoker friends during JHS + >=1 smoker friends 76.23%
during HS + norms during HS >=47.5% + Non-White + no
smoker parent (N=10)
3. Have no smoker friends during JHS + >= 1 smoker friends 76.13% 15
during HS + norms during HS <47.5% (N=31)

16. Conclusions
• Threshold of peer influences
• having 2+ smoker friends during HS gives
high probability of being smokers
• having no smoker friends during JHS gives
high probability of being non-smokers
• Threshold of parental influences
• having 1 or less smoker parent gives high
probability of being non-smokers
16

17. Limitations
• Self-reported measures
• Reduced variations of social influences
factors by using averages across waves
• Use of only perceived friend use, social
norms, and parent use to represent social
influences
17

18. Implications
• Interventions start at junior high school to
prevent students becoming cigarette
smokers
• counteract the social influences from peers
and parents
• Interventions target high-risk group
• high school students who have 2+ smoker
friends
18

19. Acknowledgments
• Funded by NIH R01DA027226 (Chou, PI)
19